[thirdparty/openssl.git] / crypto / armv4cpuid.pl

#! /usr/bin/env perl
# Copyright 2015-2023 The OpenSSL Project Authors. All Rights Reserved.
#
# Licensed under the Apache License 2.0 (the "License").  You may not use
# this file except in compliance with the License.  You can obtain a copy
# in the file LICENSE in the source distribution or at
# https://www.openssl.org/source/license.html


# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;

$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
( $xlate="${dir}perlasm/arm-xlate.pl" and -f $xlate) or
die "can't locate arm-xlate.pl";

open OUT,"| \"$^X\" $xlate $flavour \"$output\""
    or die "can't call $xlate: $!";
*STDOUT=*OUT;

$code.=<<___;
#include "arm_arch.h"

#if defined(__thumb2__) && !defined(__APPLE__)
.syntax	unified
.thumb
#else
.code	32
#undef	__thumb2__
#endif

.text

.align	5
.global	OPENSSL_atomic_add
.type	OPENSSL_atomic_add,%function
OPENSSL_atomic_add:
#if __ARM_ARCH__>=6
.Ladd:	ldrex	r2,[r0]
	add	r3,r2,r1
	strex	r2,r3,[r0]
	cmp	r2,#0
	bne	.Ladd
	mov	r0,r3
	bx	lr
#else
	stmdb	sp!,{r4-r6,lr}
	ldr	r2,.Lspinlock
	adr	r3,.Lspinlock
	mov	r4,r0
	mov	r5,r1
	add	r6,r3,r2	@ &spinlock
	b	.+8
.Lspin:	bl	sched_yield
	mov	r0,#-1
	swp	r0,r0,[r6]
	cmp	r0,#0
	bne	.Lspin

	ldr	r2,[r4]
	add	r2,r2,r5
	str	r2,[r4]
	str	r0,[r6]		@ release spinlock
	ldmia	sp!,{r4-r6,lr}
	tst	lr,#1
	moveq	pc,lr
	.word	0xe12fff1e	@ bx	lr
#endif
.size	OPENSSL_atomic_add,.-OPENSSL_atomic_add

.global	OPENSSL_cleanse
.type	OPENSSL_cleanse,%function
OPENSSL_cleanse:
	eor	ip,ip,ip
	cmp	r1,#7
#ifdef	__thumb2__
	itt	hs
#endif
	subhs	r1,r1,#4
	bhs	.Lot
	cmp	r1,#0
	beq	.Lcleanse_done
.Little:
	strb	ip,[r0],#1
	subs	r1,r1,#1
	bhi	.Little
	b	.Lcleanse_done

.Lot:	tst	r0,#3
	beq	.Laligned
	strb	ip,[r0],#1
	sub	r1,r1,#1
	b	.Lot
.Laligned:
	str	ip,[r0],#4
	subs	r1,r1,#4
	bhs	.Laligned
	adds	r1,r1,#4
	bne	.Little
.Lcleanse_done:
#if __ARM_ARCH__>=5
	bx	lr
#else
	tst	lr,#1
	moveq	pc,lr
	.word	0xe12fff1e	@ bx	lr
#endif
.size	OPENSSL_cleanse,.-OPENSSL_cleanse

.global	CRYPTO_memcmp
.type	CRYPTO_memcmp,%function
.align	4
CRYPTO_memcmp:
	eor	ip,ip,ip
	cmp	r2,#0
	beq	.Lno_data
	stmdb	sp!,{r4,r5}

.Loop_cmp:
	ldrb	r4,[r0],#1
	ldrb	r5,[r1],#1
	eor	r4,r4,r5
	orr	ip,ip,r4
	subs	r2,r2,#1
	bne	.Loop_cmp

	ldmia	sp!,{r4,r5}
.Lno_data:
	rsb	r0,ip,#0
	mov	r0,r0,lsr#31
#if __ARM_ARCH__>=5
	bx	lr
#else
	tst	lr,#1
	moveq	pc,lr
	.word	0xe12fff1e	@ bx	lr
#endif
.size	CRYPTO_memcmp,.-CRYPTO_memcmp

#if __ARM_MAX_ARCH__>=7
.arch	armv7-a
.fpu	neon

.align	5
.global	_armv7_neon_probe
.type	_armv7_neon_probe,%function
_armv7_neon_probe:
	vorr	q0,q0,q0
	bx	lr
.size	_armv7_neon_probe,.-_armv7_neon_probe

.global	_armv7_tick
.type	_armv7_tick,%function
_armv7_tick:
#ifdef	__APPLE__
	mrrc	p15,0,r0,r1,c14		@ CNTPCT
#else
	mrrc	p15,1,r0,r1,c14		@ CNTVCT
#endif
	bx	lr
.size	_armv7_tick,.-_armv7_tick

.global	_armv8_aes_probe
.type	_armv8_aes_probe,%function
_armv8_aes_probe:
#if defined(__thumb2__) && !defined(__APPLE__)
	.byte	0xb0,0xff,0x00,0x03	@ aese.8	q0,q0
#else
	.byte	0x00,0x03,0xb0,0xf3	@ aese.8	q0,q0
#endif
	bx	lr
.size	_armv8_aes_probe,.-_armv8_aes_probe

.global	_armv8_sha1_probe
.type	_armv8_sha1_probe,%function
_armv8_sha1_probe:
#if defined(__thumb2__) && !defined(__APPLE__)
	.byte	0x00,0xef,0x40,0x0c	@ sha1c.32	q0,q0,q0
#else
	.byte	0x40,0x0c,0x00,0xf2	@ sha1c.32	q0,q0,q0
#endif
	bx	lr
.size	_armv8_sha1_probe,.-_armv8_sha1_probe

.global	_armv8_sha256_probe
.type	_armv8_sha256_probe,%function
_armv8_sha256_probe:
#if defined(__thumb2__) && !defined(__APPLE__)
	.byte	0x00,0xff,0x40,0x0c	@ sha256h.32	q0,q0,q0
#else
	.byte	0x40,0x0c,0x00,0xf3	@ sha256h.32	q0,q0,q0
#endif
	bx	lr
.size	_armv8_sha256_probe,.-_armv8_sha256_probe
.global	_armv8_pmull_probe
.type	_armv8_pmull_probe,%function
_armv8_pmull_probe:
#if defined(__thumb2__) && !defined(__APPLE__)
	.byte	0xa0,0xef,0x00,0x0e	@ vmull.p64	q0,d0,d0
#else
	.byte	0x00,0x0e,0xa0,0xf2	@ vmull.p64	q0,d0,d0
#endif
	bx	lr
.size	_armv8_pmull_probe,.-_armv8_pmull_probe
#endif

.global	OPENSSL_wipe_cpu
.type	OPENSSL_wipe_cpu,%function
OPENSSL_wipe_cpu:
#if __ARM_MAX_ARCH__>=7
	ldr	r0,.LOPENSSL_armcap
	adr	r1,.LOPENSSL_armcap
	ldr	r0,[r1,r0]
#ifdef	__APPLE__
	ldr	r0,[r0]
#endif
#endif
	eor	r2,r2,r2
	eor	r3,r3,r3
	eor	ip,ip,ip
#if __ARM_MAX_ARCH__>=7
	tst	r0,#1
	beq	.Lwipe_done
	veor	q0, q0, q0
	veor	q1, q1, q1
	veor	q2, q2, q2
	veor	q3, q3, q3
	veor	q8, q8, q8
	veor	q9, q9, q9
	veor	q10, q10, q10
	veor	q11, q11, q11
	veor	q12, q12, q12
	veor	q13, q13, q13
	veor	q14, q14, q14
	veor	q15, q15, q15
.Lwipe_done:
#endif
	mov	r0,sp
#if __ARM_ARCH__>=5
	bx	lr
#else
	tst	lr,#1
	moveq	pc,lr
	.word	0xe12fff1e	@ bx	lr
#endif
.size	OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu

.global	OPENSSL_instrument_bus
.type	OPENSSL_instrument_bus,%function
OPENSSL_instrument_bus:
	eor	r0,r0,r0
#if __ARM_ARCH__>=5
	bx	lr
#else
	tst	lr,#1
	moveq	pc,lr
	.word	0xe12fff1e	@ bx	lr
#endif
.size	OPENSSL_instrument_bus,.-OPENSSL_instrument_bus

.global	OPENSSL_instrument_bus2
.type	OPENSSL_instrument_bus2,%function
OPENSSL_instrument_bus2:
	eor	r0,r0,r0
#if __ARM_ARCH__>=5
	bx	lr
#else
	tst	lr,#1
	moveq	pc,lr
	.word	0xe12fff1e	@ bx	lr
#endif
.size	OPENSSL_instrument_bus2,.-OPENSSL_instrument_bus2

.align	5
#if __ARM_MAX_ARCH__>=7
.LOPENSSL_armcap:
.word	OPENSSL_armcap_P-.
#endif
#if __ARM_ARCH__>=6
.align	5
#else
.Lspinlock:
.word	atomic_add_spinlock-.Lspinlock
.align	5

.data
.align	2
atomic_add_spinlock:
.word	0
#endif

.extern	OPENSSL_armcap_P
___

print $code;
close STDOUT or die "error closing STDOUT: $!";
Commit	Line	Data
e0a65194	1	#! /usr/bin/env perl
da1c088f	2	# Copyright 2015-2023 The OpenSSL Project Authors. All Rights Reserved.
e0a65194	3	#
0e9725bc	4	# Licensed under the Apache License 2.0 (the "License"). You may not use
e0a65194 RS	5	# this file except in compliance with the License. You can obtain a copy
	6	# in the file LICENSE in the source distribution or at
	7	# https://www.openssl.org/source/license.html
	8
313e6ec1	9
1aa89a7a RL	10	# $output is the last argument if it looks like a file (it has an extension)
	11	# $flavour is the first argument if it doesn't look like a file
	12	$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m\|\.\w+$\| ? pop : undef;
	13	$flavour = $#ARGV >= 0 && $ARGV[0] !~ m\|\.\| ? shift : undef;
313e6ec1 AP	14
	15	$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
	16	( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
	17	( $xlate="${dir}perlasm/arm-xlate.pl" and -f $xlate) or
	18	die "can't locate arm-xlate.pl";
	19
1aa89a7a RL	20	open OUT,"\| \"$^X\" $xlate $flavour \"$output\""
1aa89a7a RL	21	or die "can't call $xlate: $!";
313e6ec1 AP	22	STDOUT=OUT;
	23
	24	$code.=<<___;
87873f43 AP	25	#include "arm_arch.h"
87873f43 AP	26
11208dcf AP	27	#if defined(__thumb2__) && !defined(__APPLE__)
	28	.syntax unified
	29	.thumb
	30	#else
87873f43	31	.code 32
c93f06c1	32	#undef __thumb2__
11208dcf	33	#endif
87873f43	34
3405db97 AP	35	.text
3405db97 AP	36
4afa9f03	37	.align 5
87873f43 AP	38	.global OPENSSL_atomic_add
	39	.type OPENSSL_atomic_add,%function
	40	OPENSSL_atomic_add:
	41	#if __ARM_ARCH__>=6
	42	.Ladd: ldrex r2,[r0]
	43	add r3,r2,r1
	44	strex r2,r3,[r0]
	45	cmp r2,#0
	46	bne .Ladd
	47	mov r0,r3
5dcf70a1	48	bx lr
87873f43 AP	49	#else
	50	stmdb sp!,{r4-r6,lr}
	51	ldr r2,.Lspinlock
	52	adr r3,.Lspinlock
	53	mov r4,r0
	54	mov r5,r1
	55	add r6,r3,r2 @ &spinlock
	56	b .+8
	57	.Lspin: bl sched_yield
	58	mov r0,#-1
	59	swp r0,r0,[r6]
	60	cmp r0,#0
	61	bne .Lspin
	62
	63	ldr r2,[r4]
29fd6746	64	add r2,r2,r5
87873f43 AP	65	str r2,[r4]
	66	str r0,[r6] @ release spinlock
	67	ldmia sp!,{r4-r6,lr}
	68	tst lr,#1
	69	moveq pc,lr
	70	.word 0xe12fff1e @ bx lr
	71	#endif
	72	.size OPENSSL_atomic_add,.-OPENSSL_atomic_add
	73
	74	.global OPENSSL_cleanse
	75	.type OPENSSL_cleanse,%function
	76	OPENSSL_cleanse:
	77	eor ip,ip,ip
	78	cmp r1,#7
11208dcf AP	79	#ifdef __thumb2__
	80	itt hs
	81	#endif
29fd6746	82	subhs r1,r1,#4
87873f43 AP	83	bhs .Lot
	84	cmp r1,#0
	85	beq .Lcleanse_done
	86	.Little:
	87	strb ip,[r0],#1
29fd6746	88	subs r1,r1,#1
87873f43 AP	89	bhi .Little
	90	b .Lcleanse_done
	91
	92	.Lot: tst r0,#3
	93	beq .Laligned
	94	strb ip,[r0],#1
29fd6746	95	sub r1,r1,#1
87873f43 AP	96	b .Lot
	97	.Laligned:
	98	str ip,[r0],#4
29fd6746	99	subs r1,r1,#4
87873f43	100	bhs .Laligned
29fd6746	101	adds r1,r1,#4
87873f43 AP	102	bne .Little
87873f43 AP	103	.Lcleanse_done:
5dcf70a1 AP	104	#if __ARM_ARCH__>=5
	105	bx lr
	106	#else
87873f43 AP	107	tst lr,#1
	108	moveq pc,lr
	109	.word 0xe12fff1e @ bx lr
5dcf70a1	110	#endif
87873f43 AP	111	.size OPENSSL_cleanse,.-OPENSSL_cleanse
87873f43 AP	112
e33826f0 AP	113	.global CRYPTO_memcmp
	114	.type CRYPTO_memcmp,%function
	115	.align 4
	116	CRYPTO_memcmp:
	117	eor ip,ip,ip
	118	cmp r2,#0
	119	beq .Lno_data
	120	stmdb sp!,{r4,r5}
	121
	122	.Loop_cmp:
	123	ldrb r4,[r0],#1
	124	ldrb r5,[r1],#1
	125	eor r4,r4,r5
	126	orr ip,ip,r4
	127	subs r2,r2,#1
	128	bne .Loop_cmp
	129
	130	ldmia sp!,{r4,r5}
	131	.Lno_data:
198a2ed7	132	rsb r0,ip,#0
e33826f0 AP	133	mov r0,r0,lsr#31
	134	#if __ARM_ARCH__>=5
	135	bx lr
	136	#else
	137	tst lr,#1
	138	moveq pc,lr
	139	.word 0xe12fff1e @ bx lr
	140	#endif
	141	.size CRYPTO_memcmp,.-CRYPTO_memcmp
	142
c1669e1c AP	143	#if __ARM_MAX_ARCH__>=7
	144	.arch armv7-a
	145	.fpu neon
	146
	147	.align 5
	148	.global _armv7_neon_probe
	149	.type _armv7_neon_probe,%function
	150	_armv7_neon_probe:
	151	vorr q0,q0,q0
	152	bx lr
	153	.size _armv7_neon_probe,.-_armv7_neon_probe
	154
	155	.global _armv7_tick
	156	.type _armv7_tick,%function
	157	_armv7_tick:
313e6ec1 AP	158	#ifdef __APPLE__
	159	mrrc p15,0,r0,r1,c14 @ CNTPCT
	160	#else
c1669e1c	161	mrrc p15,1,r0,r1,c14 @ CNTVCT
313e6ec1	162	#endif
c1669e1c AP	163	bx lr
	164	.size _armv7_tick,.-_armv7_tick
	165
	166	.global _armv8_aes_probe
	167	.type _armv8_aes_probe,%function
	168	_armv8_aes_probe:
11208dcf AP	169	#if defined(__thumb2__) && !defined(__APPLE__)
	170	.byte 0xb0,0xff,0x00,0x03 @ aese.8 q0,q0
	171	#else
c1669e1c	172	.byte 0x00,0x03,0xb0,0xf3 @ aese.8 q0,q0
11208dcf	173	#endif
c1669e1c AP	174	bx lr
	175	.size _armv8_aes_probe,.-_armv8_aes_probe
	176
	177	.global _armv8_sha1_probe
	178	.type _armv8_sha1_probe,%function
	179	_armv8_sha1_probe:
11208dcf AP	180	#if defined(__thumb2__) && !defined(__APPLE__)
	181	.byte 0x00,0xef,0x40,0x0c @ sha1c.32 q0,q0,q0
	182	#else
c1669e1c	183	.byte 0x40,0x0c,0x00,0xf2 @ sha1c.32 q0,q0,q0
11208dcf	184	#endif
c1669e1c AP	185	bx lr
	186	.size _armv8_sha1_probe,.-_armv8_sha1_probe
	187
	188	.global _armv8_sha256_probe
	189	.type _armv8_sha256_probe,%function
	190	_armv8_sha256_probe:
11208dcf AP	191	#if defined(__thumb2__) && !defined(__APPLE__)
	192	.byte 0x00,0xff,0x40,0x0c @ sha256h.32 q0,q0,q0
	193	#else
c1669e1c	194	.byte 0x40,0x0c,0x00,0xf3 @ sha256h.32 q0,q0,q0
11208dcf	195	#endif
c1669e1c AP	196	bx lr
	197	.size _armv8_sha256_probe,.-_armv8_sha256_probe
	198	.global _armv8_pmull_probe
	199	.type _armv8_pmull_probe,%function
	200	_armv8_pmull_probe:
11208dcf AP	201	#if defined(__thumb2__) && !defined(__APPLE__)
	202	.byte 0xa0,0xef,0x00,0x0e @ vmull.p64 q0,d0,d0
	203	#else
c1669e1c	204	.byte 0x00,0x0e,0xa0,0xf2 @ vmull.p64 q0,d0,d0
11208dcf	205	#endif
c1669e1c AP	206	bx lr
	207	.size _armv8_pmull_probe,.-_armv8_pmull_probe
	208	#endif
	209
87873f43 AP	210	.global OPENSSL_wipe_cpu
	211	.type OPENSSL_wipe_cpu,%function
	212	OPENSSL_wipe_cpu:
c1669e1c	213	#if __ARM_MAX_ARCH__>=7
87873f43 AP	214	ldr r0,.LOPENSSL_armcap
	215	adr r1,.LOPENSSL_armcap
	216	ldr r0,[r1,r0]
313e6ec1 AP	217	#ifdef __APPLE__
	218	ldr r0,[r0]
	219	#endif
c1669e1c	220	#endif
87873f43 AP	221	eor r2,r2,r2
	222	eor r3,r3,r3
	223	eor ip,ip,ip
c1669e1c	224	#if __ARM_MAX_ARCH__>=7
87873f43 AP	225	tst r0,#1
87873f43 AP	226	beq .Lwipe_done
c1669e1c AP	227	veor q0, q0, q0
	228	veor q1, q1, q1
	229	veor q2, q2, q2
	230	veor q3, q3, q3
	231	veor q8, q8, q8
	232	veor q9, q9, q9
	233	veor q10, q10, q10
	234	veor q11, q11, q11
	235	veor q12, q12, q12
	236	veor q13, q13, q13
	237	veor q14, q14, q14
	238	veor q15, q15, q15
87873f43	239	.Lwipe_done:
c1669e1c	240	#endif
87873f43	241	mov r0,sp
5dcf70a1 AP	242	#if __ARM_ARCH__>=5
	243	bx lr
	244	#else
87873f43 AP	245	tst lr,#1
	246	moveq pc,lr
	247	.word 0xe12fff1e @ bx lr
5dcf70a1	248	#endif
87873f43 AP	249	.size OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
	250
	251	.global OPENSSL_instrument_bus
	252	.type OPENSSL_instrument_bus,%function
	253	OPENSSL_instrument_bus:
	254	eor r0,r0,r0
5dcf70a1 AP	255	#if __ARM_ARCH__>=5
	256	bx lr
	257	#else
87873f43 AP	258	tst lr,#1
	259	moveq pc,lr
	260	.word 0xe12fff1e @ bx lr
5dcf70a1	261	#endif
87873f43 AP	262	.size OPENSSL_instrument_bus,.-OPENSSL_instrument_bus
	263
	264	.global OPENSSL_instrument_bus2
	265	.type OPENSSL_instrument_bus2,%function
	266	OPENSSL_instrument_bus2:
	267	eor r0,r0,r0
5dcf70a1 AP	268	#if __ARM_ARCH__>=5
	269	bx lr
	270	#else
87873f43 AP	271	tst lr,#1
	272	moveq pc,lr
	273	.word 0xe12fff1e @ bx lr
5dcf70a1	274	#endif
87873f43 AP	275	.size OPENSSL_instrument_bus2,.-OPENSSL_instrument_bus2
	276
	277	.align 5
c1669e1c	278	#if __ARM_MAX_ARCH__>=7
87873f43	279	.LOPENSSL_armcap:
313e6ec1	280	.word OPENSSL_armcap_P-.
c1669e1c	281	#endif
87873f43 AP	282	#if __ARM_ARCH__>=6
	283	.align 5
	284	#else
	285	.Lspinlock:
	286	.word atomic_add_spinlock-.Lspinlock
	287	.align 5
	288
	289	.data
	290	.align 2
	291	atomic_add_spinlock:
	292	.word 0
	293	#endif
	294
7b508cd1	295	.extern OPENSSL_armcap_P
313e6ec1 AP	296	___
	297
	298	print $code;
a21314db	299	close STDOUT or die "error closing STDOUT: $!";